Published data is immutable β frozen ledger days and event timelines are never edited. When a value turns out to be wrong, the correction is recorded here (and served inline by /api/v1/snapshot, /api/v1/export and /api/v1/changes for the affected dates).
bulk:node_ramp:test-vehicle-2026-11-12capacity_signalsannotation on frozen history2026-11-13
TSMC A14 node ramp published as 90%, high confidence (2 readings, Jul 27 and Oct 26, 2026) β withdrawnaffected 2026-07-27 β 2026-11-13
Withdrawn: on its Q2 2026 earnings call TSMC said an internal product-like vehicle showed close to 90% device performance and close to 90% 256 Mb SRAM yield for A14. That is a development test result, not a 90% ramp of the node.
bulk:foundry_utilization:wrong-quarter-2026-11-12capacity_signalsannotation on frozen history2026-11-13
SMIC utilization for Q1 2026 published as 93.7% (12 weekly readings, Aug 17 to Oct 26, 2026) β withdrawnaffected 2026-08-17 β 2026-11-13
Withdrawn: 93.7% is the utilization SMIC reported for Q2 2026 (release of Aug 13, 2026); SMIC reported 93.1% for Q1 2026 (release of May 14, 2026). The cited analyst page puts 93.7% against Q1 2026, so every copy carried the wrong quarter.
dram-ddr4-price:Oct 2026:ddr4-8gb-spot Β· value_midmarket_datacommunity-sourced correction2026-11-13
47.107 β 46.864affected 2026-10-24 β 2026-11-13
Re-based to the series rule, the last reading of the month (owner ruling 2026-11-11). 47.107 was the Oct 23 DRAMeXchange (TrendForce) daily session average, served while October was in progress; the month's last reading is 46.864 (DRAMeXchange (TrendForce) daily session average, DDR4 8Gb 1Gx8 3200, as of Oct 30, 2026). The change is the month completing, not a restated print.
/market-data/dram-ddr4-price:unitpage_copycommunity-sourced correction2026-11-13
$/unit β $ per chip, DDR4 8Gb = 1 GBaffected 2026-03-05 β 2026-11-13
Label correction (owner ruling 2026-11-11): every DDR4 value was and is US dollars per DDR4 8Gb (1Gx8) chip; the label "$/unit" did not say which unit. One DDR4 8Gb chip holds 1 GB, so the same figures are dollars per GB. No value changed.
/tools/allocation:hbm-micron-allocation-notepage_copycommunity-sourced correction2026-11-11
HBM3E: Full production. 12-Hi yields surpassed 8-Hi. 30% lower power than competitors. HBM4: Volume production started Q1 FY2026. Meeting only 50-66% of customer demand. Record FQ2 2026: 74.9% gross margin. CapEx raised to >$25B for FY2026. Customers: 4 paying customers total. Expansion notes: Est. 20% share of 2026 HBM4 allocation. β Micron's own statements, dated: 12-high HBM3E the majority of its fiscal 2026 HBM shipments, HBM4 36GB 12-high volume production in fiscal 2026, 48GB 16-high samples, base logic and DRAM core dies made in-house (FY2026 Form 10-K, 2026-10-09); fiscal Q2 2026 revenue $23.86B with a 74.9% non-GAAP gross margin (2026-03-18). Yield, power, demand-coverage, capex and customer-count claims removed; the NVIDIA HBM4 share credited to the press report it came from (Hankyung, 2026-02-09: the provisional allocation NVIDIA made in December 2025)affected 2026-04-07 β 2026-11-11
Note correction. The yield, power and demand-coverage figures, the capex figure and the customer count were stated without a source, and the gross margin was non-GAAP without saying so. The note now quotes Micron's filings with dates and credits the NVIDIA share to the press report it came from (Hankyung, 2026-02-09).
/tools/allocation:hbm-samsung-allocation-notepage_copycommunity-sourced correction2026-11-11
HBM3E: In mass production. NVIDIA qualified Sep 2025 after 18 months of failed attempts. HBM4: World's first commercial HBM4 shipped Feb 12, 2026. Vertically integrated: 1c DRAM + 4nm foundry base die. Turnkey advantage: memory + foundry + packaging under one roof. Customers: Google (>60% of Google HBM). Expansion notes: Est. 25% share of 2026 HBM4 allocation. β Samsung's own statements, dated: HBM3E being sold to all related customers (3Q25 results); HBM4 mass production and first commercial shipments on 1c DRAM with a 4nm logic base die, and HBM sales expected to more than triple in 2026 (Samsung newsroom, 2026-02-12). The NVIDIA qualification history, the Google share, the world's-first wording in our voice and the turnkey-advantage claim removed; the NVIDIA HBM4 share credited to the press report it came from (Hankyung, 2026-02-09: the provisional allocation NVIDIA made in December 2025)affected 2026-04-07 β 2026-11-11
Note correction. The NVIDIA qualification history, the share of Google's HBM and the turnkey-advantage claim were stated without a source, and the note stated world's first as fact in our own voice; Samsung's release (2026-02-12) calls the shipment a first in the industry, by its own account. The note now quotes Samsung's results and newsroom with dates and credits the NVIDIA share to the press report it came from (Hankyung, 2026-02-09).
/tools/allocation:hbm-sk-hynix-allocation-notepage_copycommunity-sourced correction2026-11-11
HBM3E: Fully allocated through 2026+. HBM4: Mass production started Feb 2026 at M16 Icheon + M15X Cheongju. Using TSMC 12nm for base die. Dominant position. M15X started 4 months ahead of schedule. CapEx $20.5B for 2026. Customers: NVIDIA (primary). Expansion notes: Est. 55% share of 2026 HBM4 allocation. β SK hynix's own statements, dated: record 3Q25 results on 12-high HBM3E and server DDR5 (3Q25 results); HBM4 in mass production from 2025 and in mass shipments from Q2 2026 (3Q25 and 2Q26 results); FY2025 revenue KRW 97.1T with a 58% Q4 operating margin (FY2025 results); long-term memory agreements, not HBM-specific, with around 10 customers (2Q26 results). Allocation, fab, base-die and capex claims removed; the NVIDIA HBM4 share credited to the press report it came from (Hankyung, 2026-02-09: the provisional allocation NVIDIA made in December 2025)affected 2026-04-07 β 2026-11-11
Note correction. The allocation, production-site, base-die and capex claims were stated without a source, and the HBM4 date was wrong: SK hynix entered HBM4 mass production in 2025 (3Q25 results, 2025-10-29), not in February 2026. SK hynix said on 2024-04-19 that it would adopt TSMC's advanced logic process for the HBM4 base die; that release names no process node. The note now quotes SK hynix's releases with dates and credits the NVIDIA share to the press report it came from (Hankyung, 2026-02-09).
/tools/allocation:pkg-wmcm-allocation-notepage_copycommunity-sourced correction2026-11-11
Hybrid of InFO and CoWoS. Ramping to ~60K WPM by end 2026, >120K WPM by 2027. Dedicated primarily to the Apple flagship smartphone platform (public reports). Record fields: Source label 'Silicon Analysts allocation research (public sources: TSMC IR, TrendForce, trade press)', dated 2026-04-07. β TSMC's 4Q25 to 3Q26 calls do not mention WMCM, and TSMC does not comment on specific customers (3Q26 call). The >120K 2027 target and the technology description removed; the 2026 target of 60,000 wafers/month kept and labelled as a Silicon Analysts estimate, as are capacity, status and the Apple share. Record fields: Source label 'TSMC 4Q25-3Q26 earnings calls; capacity, status and Apple share: Silicon Analysts estimates from press reports', dated 2026-10-15affected 2026-04-07 β 2026-11-11
Note correction. The 60K and 120K wafers/month targets and the technology description rest on press reports, not TSMC. The record keeps 60,000 wafers/month as its own estimate for end-2026, and the note now says which figures are Silicon Analysts estimates. The source label now names the TSMC calls the note cites and says the capacity, status and Apple share are our estimates from press reports.
/tools/allocation:pkg-soic-allocation-notepage_copycommunity-sourced correction2026-11-11
TSMC VP confirmed >100% CAGR 2022-2026. 4 major companies currently using SoIC. AP7 Phase 1 targets SoIC mass production 2027. Record fields: Source label 'Silicon Analysts allocation research (public sources: TSMC IR, TrendForce, trade press)', dated 2026-04-07. β TSMC's 4Q25 to 3Q26 calls give no SoIC capacity, growth rate or customer figures; TSMC said it works with customers to meet their demand (1Q26 call, 2026-04-16). Growth-rate, user-count and AP7 claims removed. Record fields: Source label 'TSMC 4Q25-3Q26 earnings calls; capacity, status and shares: Silicon Analysts estimates from press reports', dated 2026-10-15affected 2026-04-07 β 2026-11-11
Note correction. The growth rate was credited to an unnamed TSMC vice president without a date, and the user count and AP7 schedule rest on press reports. The note now says what TSMC has said and which figures are Silicon Analysts estimates. The source label now names the TSMC calls the note cites and says the capacity, status and shares are our estimates from press reports.
/tools/allocation:pkg-cowos-allocation-notepage_copycommunity-sourced correction2026-11-11
NVIDIA holds >70% of CoWoS-L specifically. OSAT outsourcing: ~240-270K wafers/yr. Lead times >50 weeks for new orders. Record fields: Amazon/AWS (5%) and Marvell (5.5%) shares cited to no source, method 'carried from the record-level source; no per-share citation published'. Others share method 'residual to 100%; includes MediaTek at ~2% (20,000 wafers)', cited to no source. Source label 'Silicon Analysts allocation research (public sources: TSMC IR, TrendForce, trade press)', dated 2026-04-07. β TSMC's own statements, dated: advanced packaging capacity very tight, working with OSAT partners (1Q26 call); back-end capacity in shortage mode (2Q26 call); CoWoS still customers' first choice (3Q26 call); advanced packaging slightly over 10% of 2025 revenue. The CoWoS-L share and OSAT volume removed; lead time labelled as our estimate; the NVIDIA, Broadcom, AMD, Amazon/AWS and Marvell shares labelled as a 36Kr report (2025-12-04) whose NVIDIA figure cites Morgan Stanley. Record fields: Amazon/AWS (5%) and Marvell (5.5%) shares cite the same 36Kr report (2025-12-04): 50,000 and 55,000 reserved wafers of its 1 million. Others share method 'residual to 100%', with MediaTek's 20,000 reserved wafers credited to that report in the note. Source label 'TSMC 4Q25, 1Q26, 2Q26 and 3Q26 earnings calls; NVIDIA, Broadcom, AMD, Amazon/AWS and Marvell shares: 36Kr 2025-12-04, NVIDIA figure citing Morgan Stanley (report); capacity, lead time and residual share: Silicon Analysts estimates from trade reports', dated 2026-10-15affected 2026-04-07 β 2026-11-11
Note correction. The CoWoS-L share was withdrawn in an earlier repair, and TSMC gives no CoWoS capacity, OSAT volume, lead time or customer split. The note now quotes TSMC with dates, labels the NVIDIA, Broadcom, AMD, Amazon/AWS and Marvell shares as a 36Kr report (2025-12-04) whose NVIDIA figure cites Morgan Stanley, and says which figures are Silicon Analysts estimates. The Amazon/AWS and Marvell shares, served as our estimates with no source, are the wafers that report says each reserved (50,000 and 55,000) over its 1 million-wafer 2026 total, and now cite it. The MediaTek figure in the Others share's method (20,000 wafers), likewise served as our estimate, is also that report's and is now credited to it in the note. The source label now names the TSMC calls the note cites and that report, and says the capacity, lead time and residual share are our estimates from trade reports.
/tools/allocation:rapidus-2nm-allocation-notepage_copycommunity-sourced correction2026-11-11
Canon confirmed as first customer. Expansion notes: ~10 received price quotations. Japanese government largest shareholder at 11.5%. Record fields: Customers 'Canon'. Key products 'Canon test chips'. Source label 'Rapidus; TrendForce; Nikkei', dated 2026-02-27. β Rapidus's own statements, dated: a basic agreement under which Rapidus manufactures the latest model of AI semiconductors in the MN-Core series to be designed by Preferred Networks (PFN) (2025-01-08); pilot line start-up in April 2025; 2nm GAA prototyping from 2025-07-18; mass production in 2027; funding from capital increases of 424.95 billion JPY, including 167.6 billion JPY from 32 companies such as Canon (2026-06-05). Quotation-count and shareholding claims removed. Record fields: Customers 'Preferred Networks', the company Rapidus's 2025-01-08 release names. Key products 'Latest model of AI semiconductors in the MN-Core series', that release's wording. Canon, which the 2026-06-05 release lists among the 32 companies that funded Rapidus, is not listed as a customer. Source label 'Rapidus releases 2025-01-08, 2025-04-01, 2025-07-18 and 2026-06-05', dated 2026-06-05affected 2026-04-07 β 2026-11-11
Note correction. Rapidus's 2025-01-08 release names Preferred Networks (PFN) as the company whose chips it will make: Rapidus manufactures the latest model of AI semiconductors in PFN's MN-Core series. Canon is one of the 32 companies that funded Rapidus (Rapidus, 2026-06-05), and no Rapidus release this record cites names it as a customer. The count of about 10 price quotations was a remark by Rapidus president Atsuyoshi Koike reported by MONOist (2026-03-02), not a customer commitment, and the 11.5% shareholding was not traced to Rapidus. The note now quotes Rapidus with dates and says which figures are Silicon Analysts estimates. The record's customer list and key products now name PFN and the product the release names, not Canon. The source label now names the Rapidus releases the note cites, not TrendForce or Nikkei.
/tools/allocation:umc-mature-allocation-notepage_copycommunity-sourced correction2026-11-11
Expansion notes: Gradual recovery trajectory. Record fields: Utilization 78% (UMC's 4Q25 figure, since superseded). Confidence high. Source label 'UMC Q4 2025 Earnings', dated 2026-01-28. β UMC's own figures, dated: company-wide utilization 78% in 4Q25 and 85% in 2Q26, with 3Q26 guided to 90%+; 22/28nm 37% of 2Q26 revenue; 2026 capex raised to USD 2.0B (2Q26 results, 2026-07-29); 2025 22nm revenue up 93% and Fab 12i Phase 3 completed in 2025 (4Q25 results, 2026-01-28). Record fields: Utilization 85%, UMC's company-wide 2Q26 figure (2Q26 results, 2026-07-29). Confidence medium. Source label 'UMC 4Q25 results 2026-01-28; UMC 2Q26 results and board announcement 2026-07-29', dated 2026-07-29affected 2026-04-07 β 2026-11-11
Note correction. UMC's 2Q26 results supersede that outlook: utilization rose to 85% in 2Q26 with 90%+ guided for 3Q26, and UMC raised its 2026 capex to USD 2.0B. The note now quotes UMC's results with their dates. The record's utilization now carries UMC's latest reported figure, 85% for 2Q26, in place of its 4Q25 figure. Confidence is now medium, not high, because the record's capacity and allocation tier are Silicon Analysts estimates and its utilization is company-wide. The source label now names both UMC releases the note cites.
/tools/allocation:smic-7nm-allocation-notepage_copycommunity-sourced correction2026-11-11
Q4 utilization 95.7% (SMIC's company-wide figure, shown on the 7nm record). Huawei dominates for Kirin 9020 and Ascend 910C/910D (stated as fact). Plans to double 7nm capacity in 2026. Expansion notes: China targeting 7nm/5nm output fivefold increase in 2 years. Record fields: Source label 'SMIC Q4 Earnings; American Affairs', dated 2026-02-15. β SMIC's own figures, dated: 2025 revenue USD 9,327 million, up 16.2%; company-wide utilization 95.7% in 4Q25 and 93.7% in 2Q26; total capacity 1,096,500 8-inch equivalent wafers a month in 2Q26. Huawei labelled as press reports; 7nm capacity-plan claims removed. Record fields: Source label 'SMIC 4Q25 results 2026-02-10; SMIC 2Q26 results 2026-08-13; Huawei products and share: press reports', dated 2026-08-13affected 2026-04-07 β 2026-11-11
Note correction. SMIC reports utilization only for the whole company and discloses no 7nm capacity, customer or capacity plan; the doubling and fivefold claims were not traced to SMIC. The note now quotes SMIC's results with dates and says which figures are Silicon Analysts estimates. The source label now names the SMIC results the note cites, not American Affairs.
/tools/allocation:gf-specialty-allocation-notepage_copycommunity-sourced correction2026-11-11
Utilization recovered to 85-86%. Expansion notes: Malta expansion + potential new fab. Record fields: Confidence high. Source label 'GlobalFoundries FY2025 results; fab-level roll-up (12-22nm band only)', dated 2026-02-15. β GF's own figures, dated: utilization of approximately 85% for 2025 on about 2.3 million 300mm-equivalent wafers, automotive revenue up about 17% to a record USD 1.4B, silicon photonics revenue roughly doubled to over USD 200M (4Q25 call, 2026-02-11); FY2025 revenue USD 6.791B (4Q25 results); Dresden to more than one million 300mm wafers a year by end-2028 (GF, 2025-10-28 and 2026-10-09); the CHIPS award, including a new 300mm fab in Malta, as NIST lists it. Record fields: Confidence medium. Source label 'GF 4Q25 results and earnings call 2026-02-11; GF releases 2025-10-28 and 2026-10-09; NIST CHIPS award page (GlobalFoundries); fab-level roll-up (12-22nm band only)', dated 2026-10-09affected 2026-04-07 β 2026-11-11
Note correction. GF put its 2025 utilization at approximately 85%, not 85-86%, and NIST lists a new 300mm fab in Malta among the CHIPS award projects, not a potential one. The note now quotes GF and NIST with dates and says which figures are Silicon Analysts estimates. Confidence is now medium, not high, because the record's capacity and allocation tier are Silicon Analysts estimates and its utilization is GF's company-wide figure. The source label now names every document the note cites.
/tools/allocation:intel-18a-allocation-notepage_copycommunity-sourced correction2026-11-11
HVM started with Panther Lake at CES Jan 2026. Yields at ~60%. A major fabless customer reportedly evaluating for future products (trade press). Record fields: Confidence high. Source label 'Intel Q4 Earnings; KeyBanc; Reuters', dated 2026-01-22. β 18A high-volume manufacturing from late 2025 (FY2025 10-K), first 18A products released at the start of 2026 (2Q26 10-Q); no yield figure (Intel discloses none; 2Q26: yields ahead of expectations); Intel Foundry external revenue USD 293M in 2Q26, up mainly on Altera (2Q26 10-Q); 18A-P for external customers; unnamed-customer claim removed. Record fields: Confidence medium. Source label 'Intel FY2025 Form 10-K; Intel 2Q26 Form 10-Q; Intel 2Q26 earnings remarks 2026-07-23; Intel Form 8-K 2026-04-08', dated 2026-07-24affected 2026-04-07 β 2026-11-11
Note correction. The ~60% yield was an analyst estimate stated as fact, the note dated the high-volume start to the product launch, which Intel's FY2025 10-K puts in late 2025, and the unnamed-customer claim was not attributed. The note now quotes Intel's filings and remarks with dates. Confidence is now medium, not high, because the record's capacity and allocation tier are Silicon Analysts estimates. The source label now names the Intel filings and remarks the note cites, not KeyBanc or Reuters.
/tools/allocation:samsung-4nm-allocation-notepage_copycommunity-sourced correction2026-11-11
The HBM workhorse. >50% of Pyeongtaek 4nm capacity allocated to internal HBM4 logic base-die production. Expansion notes: internal HBM4 demand is the growth driver, not external foundry wins. Record fields: Utilization 80%. Customers 'Charbright' and 'Tesla'. Key products 'Charbright AI' and 'Tesla components'. Internal customer-share method 'reported >50% of Pyeongtaek 4nm allocated to internal HBM4 base dies'. Source label 'TrendForce; Seoul Economic Daily', dated 2026-03-20. β Samsung's own statements, dated: HBM4 uses a 4nm base die (2026-02-12); 2Q26 Foundry earnings before incentive-related provisions improved significantly on HBM base-die demand and strong U.S. customer orders; 2H26 plan to expand 4nm LPU and base-die sales. The over-half figure is labelled as a press report (Hankyung via TrendForce) about Pyeongtaek foundry capacity, with no node split. Record fields: No utilization figure (Samsung discloses none). Customers: Samsung HBM (internal), Tsavorite Scalable Intelligence and Tesla; key products: HBM4 base dies, Omni Processing Unit (OPU) and AI4 (FSD chip), the two outside customers labelled in the note as press reports. Internal customer-share method: not a reported 4nm figure, the report giving no node split for Samsung's Pyeongtaek foundry capacity. Source label 'Samsung Newsroom 2026-02-12; Samsung 1Q26 and 2Q26 results; Hankyung via TrendForce 2026-03-20 (report); Tsavorite customer: Reuters 2025-11-10 (report); Tesla customer: heise online 2025-07-28 and KED Global 2021-09-23 (report)', dated 2026-07-30affected 2026-04-07 β 2026-11-11
Note correction. Samsung discloses no per-node allocation. The over-half figure is a press report about Samsung's Pyeongtaek foundry capacity, which the note stated as fact and narrowed to 4nm, and Samsung itself names U.S. customer orders alongside base-die demand. The 80% utilization had no source: Samsung discloses no 4nm utilization and had planned full utilization of its advanced-node lines for 2Q26, so the record now gives none. The internal customer share's method now says it is not a reported 4nm figure and that the report gives no node split for Pyeongtaek foundry capacity. 'Charbright' matches no company: it appears to be a machine rendering of the Korean name of Tsavorite Scalable Intelligence, which placed its 4nm production with Samsung (Smart Today, 2025-11-11). Tsavorite is using Samsung's SF4X 4nm-family process to fabricate its Omni Processing Unit (Reuters, 2025-11-10), and a Samsung foundry vice president said Samsung is applying SF4X to it (Smart Today, 2025-11-11), so the record names Tsavorite and the OPU instead of 'Charbright' and 'Charbright AI'. Tesla stays: Samsung is producing Tesla's AI4 chip, formerly HW4, on 4nm (heise online, 2025-07-28), although an earlier report put that chip on Samsung's 7nm process (KED Global, 2021-09-23); the note gives both reports, and the AI4 replaces the unspecific 'Tesla components'. The source label now names the documents the note cites, not Seoul Economic Daily.
/tools/allocation:samsung-3nm-allocation-notepage_copycommunity-sourced correction2026-11-11
Commercial disappointment. Yields stuck at ~50% vs TSMC >90%. Expansion notes: No new customers expected. Record fields: Customer 'Tesla'. Key product 'Tesla AI5'. Source label 'TrendForce; DigiTimes', dated 2025-12-15. β Samsung's own statements: first 3nm GAA shipment on 2022-07-25; Exynos 2500 on 3nm GAA. The S3 conversion labelled as a 2024 press report that expected it by end-2025 (Business Korea via TrendForce, 2024-10-04); the yield comparison and customer outlook removed. Record fields: Customers: Samsung (internal) only; key products: Exynos 2500 only. Source label 'Samsung Semiconductor newsroom 2022-07-25; Samsung Semiconductor Exynos 2500 page; Samsung 1Q26 and 2Q26 results; Business Korea via TrendForce 2024-10-04 (report)', dated 2026-07-30affected 2026-04-07 β 2026-11-11
Note correction. Neither Samsung nor TSMC discloses these yields, and the characterization and customer outlook were stated as fact without a source. The note now cites Samsung and labels the one press report kept. No dated report from a reputable outlet and no Samsung or Tesla statement ties Tesla to Samsung 3nm: SamMobile (2025-11-25 and 2026-04-16) says the AI5 is made on 3nm without a source for it, DigiTimes (2025-10-29) pairs the AI5 order with Samsung's 3nm yield gains in its headline and does not say in its unpaywalled text that the AI5 is made on 3nm, and the other reports of Samsung's AI5 order name no Samsung node, name a line rather than a process (the Hwaseong S3 line, Korea JoongAng Daily, 2025-11-26), give a 2-3nm range, or put it on Samsung's 2nm process (The Bell via TrendForce, 2025-11-05; Financial News, 2026-07-13). Tesla and the Tesla AI5 are removed. The source label now names the documents the note cites, not DigiTimes.
/tools/allocation:samsung-2nm-allocation-notepage_copycommunity-sourced correction2026-11-11
Yields at 50-60%. MPW delayed ~6 months. Expansion notes: Taylor TX at >90% construction. Samsung targeting 20% foundry market share by 2027. Record fields: Confidence high. Source label 'Samsung Q4 Earnings; TrendForce; Electrek', dated 2026-01-28. β Samsung's own statements, dated: 2nm mass production from 4Q25 (4Q25 results); 2nm HPC engagements (2Q26 results); a KRW 22.76 trillion foundry contract to 2033 with Tesla (DART, 2025-07-28 and 2025-07-31); Taylor operations to begin in 2026 (3Q25 results). Yield, delay, construction and share-target claims removed; the S3 conversion labelled as a 2024 press report. Record fields: Confidence medium. Source label 'Samsung 3Q25, 4Q25, 1Q26 and 2Q26 results; Samsung DART filings 2025-07-28 and 2025-07-31; Samsung Semiconductor Exynos 2600 page; Business Korea via TrendForce 2024-10-04 (report)', dated 2026-07-30affected 2026-04-07 β 2026-11-11
Note correction. Samsung discloses no 2nm yield, and the delay, construction-progress and market-share figures were not traced to Samsung. The note now quotes Samsung's results and filings with dates and labels the one press report kept. Confidence is now medium, not high, because the record's capacity, customer shares and allocation tier are Silicon Analysts estimates. The source label now names the documents the note cites, not Electrek.
/tools/allocation:tsmc-mature-allocation-notepage_copycommunity-sourced correction2026-11-11
Sub-80% utilization. Fab 14 mature capacity being cut 15-20% by 2028. JASM P1 at ~50% utilization. Expansion notes: Structural decline. Fab 14 capacity being cut. JASM P1 delaying equipment adds. Record fields: Source label 'TSMC Q4 Earnings; Counterpoint; TrendForce', dated 2026-01-22. β TSMC's own statements, dated: nodes above 7nm were 23% of 4Q25 and 21% of 3Q26 wafer revenue; mature capacity increasing in higher value-added segments such as JASM Fab 1 and ESMC (2Q26 call); JASM in volume production since late 2024 (4Q25 call). Utilization and Fab 14 cut claims removed. Record fields: Source label 'TSMC 4Q25, 2Q26 and 3Q26 earnings calls; TSMC FY2025 Form 20-F', dated 2026-10-15affected 2026-04-07 β 2026-11-11
Note correction. TSMC's calls give no mature-node utilization figure and no Fab 14 capacity cut, the JASM equipment delay was not traced to TSMC, and TSMC says it is increasing, not decreasing, mature-node capacity in higher value-added segments. The note now quotes TSMC with dates and says which figures are Silicon Analysts estimates. The source label now names the documents the note cites, not Counterpoint or TrendForce.
/tools/allocation:tsmc-n7-allocation-notepage_copycommunity-sourced correction2026-11-11
Utilization dropped below 70% as customers migrate to N5/N3. TSMC implemented 10% price cut Jan 2025. Record fields: Source label 'TrendForce; TSMC Q4 Earnings', dated 2025-12-15. β TSMC's own N7 statements, dated: 14% of 4Q25 and 10% of 3Q26 wafer revenue; flexible capacity support among N7, N5 and N3; 3nm for the second Japan fab (2028). Utilization and price-cut claims removed; the 65% in the capacity check labelled as an assumption. Record fields: Source label 'TSMC 4Q25, 1Q26 and 3Q26 earnings calls; TSMC 2Q26 IR (revenue mix x wafer shipments); TSMC FY2025 Form 20-F', dated 2026-10-15affected 2026-04-07 β 2026-11-11
Note correction. TSMC gives no N7 utilization figure, and the price cut was not traced to TSMC. The note now quotes TSMC with dates and says which figures are Silicon Analysts estimates. The source label now names the documents the note cites, not TrendForce.